Sex/Gender Differences in Cognitive Trajectories Vary as a Function of Race/Ethnicity

Abstract

INTRODUCTION:

The current study sought to determine whether cognitive trajectories differ between men and women across and within racial/ethnic groups.

METHODS:

Participants were 5,258 non-Hispanic White (NHW), Black, and Hispanic men and women in the Washington/Hamilton Heights Inwood Columbia Aging Project (WHICAP) who were administered neuropsychological tests of memory, language, and visuo-spatial abilities at 18 to 24 month intervals for up to 25 years. Multiple-group latent growth curve modeling examined trajectories across sex/gender by race/ethnicity.

RESULTS:

After adjusting for age and education, the largest baseline differences were between NHW men and Hispanic women on visuo-spatial and language, and between NHW women and Black men on memory. Memory and visuo-spatial decline was steeper for Black women compared to Hispanic men and NHW women, respectively.

DISCUSSION:

This study takes an important first step in understanding interactions between race/ethnicity and sex/gender on cognitive trajectories by demonstrating variability in sex/gender differences across race/ethnicity.

1. Background

Accurate identification of cognitive aging disparities is an important step in eventually eliminating disparities in Alzheimer’s disease (AD). Older Blacks and Hispanics are approximately 2 to 3 times more likely than Non-Hispanic Whites (NHWs) to have AD [1-4]. Women are two-thirds of the population over age 65 and represent nearly two-thirds of the 5.3 million individuals aged 65 years and older with AD [5]. Racial/ethnic differences in cognitive test performance and rate of decline have been examined, as well as between men and women, however, little has been done to understand how race/ethnicity and sex/gender intersect to define the cognitive health of older Americans. Testing the interactions of race/ethnicity with sex/gender on cognitive trajectories may yield a more nuanced understanding of mechanisms of AD disparities and lead to the development of new strategies to prevent or slow AD-related cognitive decline.

Race/ethnicity is not a reliable indicator of biological variation within the human population [6]. Rather, racial/ethnic categories serve as proxies for socio-cultural forces that shape individual level environments and resources (e.g., material conditions, psychological stressors, cognitive engagement, test taking skills), biological mediators (e.g., physical health, health behaviors), and ultimately cognition [7]. Several prior studies have documented cross-sectional differences in cognitive test performance across race/ethnicity [8-10], but findings regarding differences in rates of cognitive decline have been inconsistent [11-15].

Sex refers to biological differences between males and females (i.e., chromosomal, gonadal, or hormonal), while gender refers to socio-culturally constructed characteristics of women and men [16]. Differences between men and women on cognitive outcomes may be due to both biological and socio-cultural factors. Research has consistently demonstrated an advantage for women on episodic memory tests and an advantage for men on visuo-spatial functioning [17-19]. Research on sex/gender differences in verbal abilities in older adults is inconsistent, with some studies showing an advantage for women [17, 20, 21] and others showing no sex/gender difference [22]. Some studies have demonstrated that women demonstrate a faster rate of cognitive and functional decline after receiving a diagnosis of AD [23, 24], yet most of the literature suggests that men and women decline at similar rates [16, 25]. Sex/gender differences in children are more pronounced at higher socioeconomic status (SES) levels and in cultures with wider sex/gender stratification (i.e., roles of men and women are controlled into distinct spheres and duties) [26].

Each of the aforementioned studies on sex/gender differences are based on samples that are primarily NHW. It is unclear whether these differences are present in racial/ethnic minority populations. However, racial/ethnic differences in SES and norms/expectations about women’s work [27] suggests that sex/gender may have a different effect on cognition in Blacks and Hispanics. To date, no known studies have examined interactions between race/ethnicity and sex/gender on cognitive trajectories to AD.

The overall goal of the current study was to highlight the importance of examining the multiplicative influence of race/ethnicity and sex/gender on cognitive trajectories by leveraging longitudinal data from a community-based cohort of NHW, Black, and Hispanic older adult men and women. We examined whether initial cognitive test performance and rates of change differed across sex/gender, racial/ethnic, and sex/gender by racial/ethnic groups. We hypothesized that patterns of racial/ethnic and sex/gender differences in test performance would reflect previous findings. Namely, NHWs would obtain higher scores than Blacks and Hispanics, women obtain higher scores on memory, men obtain higher scores on visuo-spatial abilities, and no sex/gender differences on language measures. When performance was examined across sex/gender by racial/ethnic groups, we expected sex/gender differences to vary as a function of racial/ethnic identity, with sex/gender differences between NHW men and women and not Black and Hispanic men and women. We determined whether differences persisted after adjusting for differences in age, education, and cardiovascular disease burden. This study represents the first in a series of steps that reflect a three-phased disparities research framework [28] to 1) detect, 2) understand, and 3) reduce/eliminate sex/gender by racial/ethnic differences in cognitive outcomes. The detection phase requires accurate quantification of group differences that will inform subsequent steps. Comprehensive examination of potential mediators and moderators underlying these differencesis the next step in this work, but beyond the scope of the current paper.

2. Methods

2.1. Participants

Participants were community-living Medicare recipients 65 years and older recruited from northern Manhattan to participate in the Washington Heights-Inwood Columbia Aging Project (WHICAP) [29]. Potential participants were identified based on residence in U.S. census tracts within the study catchment area. Recruitment occurred in three waves: 1992 (N = 2,126), 1999 (N = 2,174) and 2009 (N = 2,128). Participants completed a baseline assessment and were followed up at 18 to 24-month intervals for up to 25 years. During each session, participants were administered a neuropsychological battery and asked about their general health, functional ability and medical history. Evaluations were conducted in English or Spanish, based on language preference. This study was approved by Institutional Review Boards at Columbia Presbyterian Medical Center, Columbia University Health Sciences, and the New York State Psychiatric Institute.

Participants who self-reported their primary race/ethnicity to be NHW, Black, or Hispanic were included in the current sample. Of the 6,639 participating individuals, 6,179 completed at least one full neuropsychological battery. Participants who self-reported a primary race/ethnicity other than NHW, Black, or Hispanic (N = 85), were missing years of education (N = 20), or who met criteria for dementia at baseline (N = 621) were excluded from the current analyses. Diagnosis of all-cause dementia was determined via consensus case conference based on neurological, neuropsychological, functional, medical, and psychiatric data collected from participants and/or informants, and followed standard research criteria for all-cause dementia [30, 31]. Follow-up diagnoses were made blind to prior diagnoses. The remaining sample (N = 5,258) included 530 NHW men, 812 NHW women, 483 Black men, 1,127 Black women, 714 Hispanic men, and 1,592 Hispanic women. Approximately 60% of the participants were administered study protocols in English. Spanish was the preferred language of administration for 94% of the Hispanic participants.

2.2. Measures

2.2.1. Demographic Measures

Self-reported race/ethnicity was classified based on the 1990 US Census guidelines. Participants were first asked whether they were Hispanic or Latino and then asked to classify themselves racially as White, Black, Asian, American Indian, Pacific Islander, or other. Participants were asked whether they are male or female; however, this method does not allow us to know whether participants reported their sex or gender [32]. Thus, we will use the term “sex/gender”. Highest completed grade of school was used for educational attainment. Self-reported history of hypertension, diabetes, heart disease, and stroke was used to create a cardiovascular disease risk score by adding the number of conditions endorsed (0-4) [33].

2.2.2. Neuropsychological Assessments

Cognitive functioning was assessed via a comprehensive neuropsychological battery (see Stern et al., 1992, for details on battery development). Confirmatory factor analysis (CFA) was conducted on this battery summarizing it into three cognitive domains: memory, language, and visuo-spatial functioning. The memory domain consisted of immediate, delayed, and recognition trails from the Selective Reminding Test (SRT) [34]. Language was measured via confrontation naming, letter and category fluency, and verbal abstract reasoning. The visuo-spatial domain consisted of recognition and matching trials from the Benton Visual Retention Test (BVRT) [35], the Rosen Drawing Test [36], and the Identities and Oddities subtest of the Mattis Dementia Rating Scale [37].

2.3. Statistical Analyses

SPSS 24 was used to generate descriptive statistics and conduct group comparisons. Mplus version 7.4 [38] was used to estimate all models using maximum likelihood estimation. Time was parameterized as years from study entry and the time score option in Mplus was used to accommodate individual differences in intervals between study visits.

The first analytic step was to determine whether cognitive test performance could be meaningfully compared across groups. A series of measurement invariance analyses were conducted to ensure construct comparability across sex/gender, racial/ethnic, and sex/gender by racial/ethnic groups, as well as across repeated measurements. Results from invariance analyses indicated that at least partial scalar invariance held across all group combinations and across five measurement occasions [39], suggesting that differences in cognitive test performance are not due to construct-irrelevant factors (e.g., lack of cultural and language equivalence). These results do not rule out the role that socio-cultural factors play in test performance differences.

Each of the cognitive variables were then converted to z-scores using means and standard deviations from the entire sample at baseline. Composite scores were computed by averaging the z-scores within each of the three domains at each occasion. Scores were then corrected for age by regressing baseline composite scores on baseline age separately for each cognitive domain. The constant and slope from each regression were used to adjust baseline and follow-up composites. Age-corrected scores of zero indicate performance equal to what would be expected for the given baseline age.

Longitudinal changes within each cognitive domain over five study waves (approximately 15 years in the study) were estimated. Missing data were handled using full information maximum likelihood (FIML). Model fit was assessed using the Bayesian Information Criterion (BIC) [40]. Relative model fit, as assed by the BIC, is based on a log-likelihood value that rewards better model fit while penalizing the presence of more model parameters. Models with a higher log-likelihood value and few parameters will have smaller BIC values [41].

Cognitive trajectories for each domain were characterized by estimating three separate unconditional latent growth curve models, specified to include no covariates, across the entire sample. Unconditional latent growth curve models that allowed only linear change were compared to models allowing both linear and quadratic change. A spline modeling retest was included if evidence of practice effects was noted [42]. The best fitting models based on BIC were used for subsequent analyses.

For each cognitive domain, three separate multiple-group models were estimated to examine intercept and slope differences between sex/gender, racial/ethnic, and sex/gender by racial/ethnic groups. The multiple-group approach is preferable to treating sex/gender and/or race/ethnicity as covariates, which would impose parameter equalities between groups that may not be valid [43].

Years of school and cardiovascular disease burden were included in conditional models to identify group differences and individual changes in cognition above and beyond those expected based on education and/or co-morbid cardiovascular conditions. Education was centered at 10 years of school.

Additional sensitivity analyses were conducted to assess whether attrition due to death and terminal decline in cognitive functioning influenced our findings. First, joint modeling that combined a latent growth model with a discrete-time survival model was used to examine attrition due to death. In the discrete-time survival model we estimated a latent hazard function representing the conditional probability (i.e., “hazard”) of death at a specific time point, given survival at an earlier time point [44]. The latent hazard function was then regressed on growth trajectories for each cognitive domain. Second, to remove the potential influence of terminal decline, all data points within 3 years of death were excluded.

3. Results

Women in the study were older and completed fewer years of education than men. NHW women had the lowest cardiovascular disease risk scores, followed by NHW and Black men who had lower scores than Black women and Hispanic men and women. Estimated Mild Cognitive Impairment (MCI) at baseline was similar across sex/gender by racial/ethnic groups, ranging from 18% in NHW men to 23% in Hispanic women. Rates of incident dementia ranged from 6% in NHW men and women to 19% in Hispanic women. Consistent with recent literature among racial/ethnic groups in the U.S. [45], there were no within racial/ethnic group sex/gender differences in incidence rates for NHWs and Blacks. However, higher rates in Hispanic women compared to men is a unique finding and may reflect cultural variability in sex/gender roles between U.S. born NHWs and Blacks and foreign-born Hispanics in our sample. Sample characteristics are provided (Table 1).

Table 1.

Sample Characteristics Across Sex/Gender by Racial/Ethnic Group

Characteristics	NHW Men (n = 530)	NHW Women (n = 812)	Black Men (n = 483)	Black Women (n = 1127)	Hispanic Men (n = 714)	Hispanic Women (n = 1592)
Age, mean (SD)	75.3 (6.3)	76.4 (7.2)	74.6 (6.4)	76.2 (6.6)	75.7 (6.1)	75.7 (6.2)
Education, mean (SD)	14.3 (3.9)	13.7 (3.5)	11.3 (3.9)	11.8 (3.5)	7.2 (4.4)	6.9 (4.4)
CVD Count, mean (SD)	0.97 (0.9)	0.84 (0.8)	1.09 (0.9)	1.27 (0.9)	1.31 (1.0)	1.32 (0.9)
Hypertension, No (%)	364 (69)	529 (65)	338 (70)	946 (84)	548 (77)	1345 (85)
Diabetes, No (%)	91 (17)	108 (13)	106 (22)	327 (29)	250 (35)	556 (35)
Heart Disease, No (%)	249 (47)	278 (34)	145 (30)	410 (36)	275 (39)	493 (31)
Stroke, No (%)	11 (4)	23 (5)	19 (7)	37 (6)	19 (6)	45 (6)
Diagnostic Status, No. (%)
MCI at Baseline	94 (18)	149 (18)	91 (19)	200 (18)	149 (21)	360 (23)
Incident Dementia	31 (6)	53 (6)	50 (10)	143 (13)	111 (11)	299 (19)
Baseline Factor Scoresa
Memory, mean (SD)	0.40 (0.7)	0.52 (0.7)	0.03 (0.7)	0.22 (0.7)	−0.08 (0.6)	0.04 (0.6)
Language, mean (SD)	0.75 (0.7)	0.68 (0.8)	0.25 (0.7)	0.16 (0.7)	−0.12 (0.6)	−0.16 (0.6)
Visuo-Spatial, mean (SD)	0.59 (0.4)	0.58 (0.4)	0.24 (0.5)	0.22 (0.6)	−0.10 (0.7)	−0.21 (0.7)

Abbreviations: NHW, Non-Hispanic White; SD, Standard Deviation; CVD, cardiovascular disease.

^aAverage baseline scores for the current sample are greater than zero because scores were normalized on the entire cohort that included individuals demented at baseline.

3.1. Characterizing Cognitive Trajectories

Unconditional models allowing only linear change fit better than the models allowing both linear and quadratic slopes. Evidence of practice effects were noted for the language model but not the memory or visuo-spatial models, so a spline for retest was only included in the language model. Average baseline scores ranged from 0.138 to 0.165 across the cognitive domains. Rates of change were significant for all cognitive domains, suggesting a decline in cognitive performance over time.

3.2. Differences Between Groups

Estimated trajectories for the unadjusted scores for each model are presented in Figure 1. Results from the unconditional age-adjusted and conditional age-adjusted multiple-group analyses are discussed below.

Figure 1.

Estimated Memory, Language, and Visuo-Spatial Trajectories for the Unadjusted Multiple-Group Models Across Sex/Gender, Racial/Ethnic, and Sex/Gender by Racial/Ethnic Groups

Sex/gender differences were noted for baseline performance across all three cognitive domains. Women demonstrated higher average baseline memory scores than men and men obtained higher scores on the language and visuo-spatial domains. Rates of decline were similar across sex/gender groups for all three domains.

Baseline performance differed between all racial/ethnic groups. For each cognitive domain, NHWs had the highest baseline scores, followed by Black participants who had higher baseline scores compared to Hispanics. Compared to NHWs, Blacks demonstrated steeper rates of decline on the memory and visuo-spatial domains. NHW and Hispanic participants demonstrated similar rates of decline across the three cognitive domains.

Differences in baseline performance and rate of change were noted on the sex/gender by race/ethnicity models. Sex/gender differences in baseline memory performance were found within each racial/ethnic group, with women obtaining higher baseline memory scores. While NHW women demonstrated the highest baseline memory performance, NHW men obtained higher scores than Black (B = 0.16; 95% CI = [0.09, 0.23]) and Hispanic women (B = 0.38; 95% CI = [0.32, 0.45]). Baseline scores were similar for Hispanic women and Black men. Memory decline was steeper for Black women compared to NHW women (B = 0.02; 95% CI = [0.01, 0.03]). Sex/gender differences were found in the visuo-spatial domain for Hispanics, with higher baseline scores for men than women (B = 0.09; 95% CI = [0.03, 0.15]). Rate of language decline was similar across sex/gender by racial/ethnic groups, however, decline for Black women on visuo-spatial functioning was steeper compared with NHW women (B = 0.01; 95% CI = [0.01, 0.02]). Intercept and slope estimates are presented in supplementary Table 2.

After adjusting for education, baseline memory differences in growth trajectories remained between NHW women and all other groups, as well as between Black and Hispanic men and all other groups. While differences in memory decline between NHW and Black women dissipated, the average memory slope for Black women at 10 years of education was steeper than that of Hispanic men (B = 0.02; 95% CI = [0.01, 0.04]). For baseline language and visuo-spatial performance, differences only remained between the NHWs and all other groups. When cardiovascular disease risk was added to each model, memory decline for Black men was steeper than Hispanic men (B = 0.03; 95% CI = [0.06, 0.01]) and there were no differences in baseline memory between NHW women and men. Compared to NHW women, baseline memory was 0.4 and 0.3 times lower from Black and Hispanic men, respectively, and memory decline was approximately 0.5 times faster for Black men and women. Intercept and slope differences across groups for each cognitive domain are summarized (Figure 2).

Figure 2.

Intercept and Slope Means and Confidence Intervals for the Unconditional Age-Adjusted and Conditional Age-Adjusted Models Across Sex/Gender by Racial/Ethnic Groups on Each Cognitive Domain

3.3. Sensitivity Analyses

Results from the joint discrete time survival and growth model indicated that risk of death was related to memory performance but not performance on the language and visuo-spatial domains. Risk of death was higher for individuals with lower baseline memory scores (hOR = −0.54 [−0.68, −0.40]). Decrease in slope was also associated with risk of death (hOR = −11.28 [−14.83, −7.73]), and the estimated slope for the memory joint model (B = −0.030; SE =.003) was steeper than the slope estimated in the conditional model (B = −0.024; SE = .002). Separate joint models were estimated for each sex/gender by racial/ethnic group. Increased risk of death was associated with lower baseline memory scores for the NHW men (hOR = −0.60 [−0.971, −0.23]), NHW women (hOR = −0.93 [−1.27, −0.39]), Black men (hOR = −0.48 [−0.93, −0.02]), and Black women (hOR = −0.55 [−0.81, −0.29]), as well as decrease in slope of NHW women (hOR = −5.64 [−11.27, −0.17]), Black men (hOR = −13.18 [−21.90, −4.46]), and Black women (hOR = −8.14 [13.49, −2.79]). Findings were identical to the conditional model regarding group differences in memory performance.

Additional sensitivity analyses were conducted that excluded data points within 3 years of death for all participants. Approximately 7% of data points were omitted. Participants with omitted data points tended to be older at baseline, have less education, and have lower cognitive scores at each time point. Overall findings were substantively unchanged to the conditional model analyses, although omission of data points within 3 years of death led to more positive memory slopes across sex/gender by racial/ethnic groups. The largest changes in slope were for NHW women (−.030 to −.020), Black men (−.027 to −.019), and Black women (−.037 to −.030).

4. Discussion

Examination of sex/gender differences supported previous literature demonstrating a baseline advantage for women in verbal memory and for men in visuo-spatial skills [17-19], however, these differences varied by race/ethnicity in our study. The sex/gender by race/ethnicity model demonstrated differences in rates of decline on memory and visuo-spatial abilities that were not detected when looking at sex/gender and race/ethnicity separately.

Researchers have long believed that men have a visuo-spatial advantage through late adulthood that is linked to early organizational and later activational effects of sex steroid hormones [46]. The advantage of women over men on tests of verbal memory has been documented throughout the lifecourse [25]. Studies have demonstrated that this advantage is maintained during the prodromal stage of AD [47], suggesting a greater resilience or cognitive reserve that protects from age-related cognitive decline in older women. None of the aforementioned studies examined whether these sex/gender differences varied across race/ethnicity.

In the current study, men obtained higher scores than women on the language and visuo-spatial domains, but when performance was examined across sex/gender by race/ethnicity groups only Hispanic men outperformed their female counterparts on the visuo-spatial domain, while NHW and Black men performed similarly to their female counterparts. After adjusting for education, the visuo-spatial advantage was only demonstrated for NHW men compared with Black and Hispanic women. Black and Hispanic women had a relative advantage over Black and Hispanic men on memory performance, respectively, but NHW women obtained higher scores than all sex/gender by racial/ethnic groups. These findings are important because most research on sex/gender differences has been conducted in primarily NHW samples.

Research on differences in rates of cognitive decline between race/ethnicity and sex/gender has been inconsistent, particularly across sex/gender [25]. In the current study, racial/ethnic differences in rate of memory and visuo-spatial decline were noted between the NHW and Black groups. When these differences were further teased apart in the sex/gender by race/ethnicity model, Black women demonstrated a steeper decline compared with NHW women, while NHW and Black men tended to decline at similar rates. When we accounted for educational differences between groups by centering years of education at the sample mean (10 years), average rates of memory decline for NHW women became more negative and similar to rates of decline for Black women.

For Hispanic men, adjusting for education resulted in an average rate of memory decline that was less steep compared with Black women. This difference remained after excluding data points within three years of death. Black men demonstrated the steepest memory slopes in the joint model. These findings suggest that the conditional latent growth curve models may underestimate rate of memory decline across groups because individuals who remained alive and participated tended to demonstrate less steep declines in functioning, especially Black men. Similarly, results from the sensitivity analyses suggest that memory trajectories may be sensitive to terminal decline.

Most research on sex/gender differences in cognitive test performance, cognitive change, and risk for AD assumes that biology, especially hormonal influences aacross the lifespan [48, 49] drives differenes between men and women. Our finding that sex/gender differences were not universal across racial/ethnic groups suggests that biological and hormonal factors should not be the sole focus of this research.

The sample was recruited from nm residents, which is a limitation for national generalizability. In particular, Hispanics in nm are primarily Caribbean emigrants and it is unclear if these results apply to other Hispanic subgroups who have considerably different cultural, immigration, and educational experiences.

In most cognitive aging research, the effects of race/ethnicity are typically examined while stratifying by or controlling for sex/gender, or vice versa. This approach fails to recognize potential racial/ethnic-related variability in sex/gender groups in cognitive trajectories leading to AD. Overall, our findings suggest that sex/gender differences in cognitive performance vary as a function of race/ethnicity. This study takes an important first step in understanding the simultaneous influence of race/ethnicity and sex/gender on cognitive trajectories. An important subsequent step involves accurately identifying the causal mechanisms underlying these differences and examining sex/gender-related variability in AD-risk factors across and within racial/ethnic groups.

Research in Context.

SYSTEMATIC REVIEW: Literature was reviewed using traditional sources (e.g., PsycINFO, PubMed). Since no known studies have examined interactions between race/ethnicity and sex/gender on cognitive trajectories to Alzheimer’s disease (AD), research describing sociocultural differences between sex/gender by racial/ethnic groups was used to inform hypotheses in the current study.

INTERPRETATION: Our findings suggest that frequently documented sex/gender differences in memory, language, and visuo-spatial test performance vary as a function of race/ethnicity.

FUTURE DIRECTIONS: This study takes an important first step in understanding interactions between race/ethnicity and sex/gender on cognitive trajectories. Additional studies are warranted to understand the mechanisms underlying sex/gender by race/ethnicity differences. Examples include: (a) accurately identifying the socio-cultural mechanisms underlying these differences; (b) examining the interplay between biological and socio-cultural mechanisms contributing to these differences; and (c) investigating sex/gender-related variability in AD and AD-risk factors across and within racial/ethnic groups.